A Deep Learning Approach for Rapid and Cost-Effective Detection of Pesticide Residues in Agricultural Products

Detecting pesticide residues in agricultural products is critical for ensuring food safety and promoting environmental sustainability. This study proposes a deep learning-based image processing approach for the rapid and cost-effective detection of pesticide residues. For this purpose, images of grape samples were used to classify fluxapyroxad residues into three classes ‘low’, ‘medium’, and ‘high’. In the image pre-processing stage, the automatic ROI (Region of Interest) extraction method was applied to better capture pesticide-induced visual anomalies. ResNet50 and EfficientNetB3 architectures were used in the model training process and their performance was compared. The ResNet50 model showed the best performance with an overall accuracy of 83.17%, while the EfficientNetB3 model achieved a recall of 85%, especially for samples with high pesticide concentrations. The results show that deep learning-based models can provide faster and more economical pesticide detection compared to traditional methods. In this context, it was assessed that the proposed approach can be integrated into agricultural quality control processes.